Lists
Lists
Lohiya, Krutika, Thakare, Anuradha, Gudadhe, Santwana, Virdee, Bal Singh, Khanna, Ashish, Bokefode, Pratham, Thakur, Isha, Dighole, Shamal and Ajmera, Lokesh (2026) AIMRS: an explainable optimized multi-output machine learning framework for personalized ayurvedic formulation, herb, diet, and yoga recommendation. Journal of Intelligent Decision Making and Information Science, 3 (12). pp. 1269-1290. ISSN 3079-0875
Ayurveda is a comprehensive personalized healthcare approach where treatment recommendations are made not only based on disease diagnosis but also based on numerous individual-specific factors like symptoms, dosha, prakriti, diet, life style, sleep habit, stress, environmental exposure, medical history, and overall body condition of the patient. The traditional approach in Ayurveda depends largely on expert knowledge and manual analysis of interrelated factors, thus resulting in an expensive, subjective, and hard-to-standardize decision-making process. While there have been some efforts in recent times using advancements in artificial intelligence in disease prediction, prakriti classification, and recognition of medicinal plants in the literature, most of these efforts consider only one particular aspect of personalized Ayurvedic recommendation and are oblivious of the multi-dimensional nature of the problem. To bridge this gap, the Ayurvedic Intelligent Multi-output Recommendation System (AIMRS), a comprehensible and optimized machine learning model for personalized Ayurvedic recommendation support. The suggested framework predicts nine recommendation targets: Ayurvedic Herbs, Formulation, Diet and Lifestyle Recommendations, Yoga and Physical Therapy, Medical Intervention, Prevention, Prognosis, Complications, and Patient Recommendations, based on the Kaggle dataset with 446 preprocessed patient records and 19 input features and 9 target outputs. In order to address issues related to sparseness, imbalance, and length of the text-based recommendation labels, the framework implements label normalization, rare class processing, semantic clustering, target-wise modeling selection, hyperparameter tuning, Top-K recommendations, ablation analysis, and explainability.The experimental evaluation of the proposed model shows that it achieves the maximum Weighted F1-score of 0.7350 on the Medical Intervention target and significant improvements on other recommendation targets following the target-space optimization. The obtained results prove that the framework of optimized target-wise modeling is efficient and explainable enough for Ayurvedic multi-output recommendations. The suggested framework is aimed at academic research and practitioner-assisted decision making and is not intended for medical diagnoses or Ayurvedic treatment.
Available under License Creative Commons Attribution Non-commercial No Derivatives 4.0.
Download (1MB) | Preview
 |
View Item |